Zuniga, E., Milani, S., Leroy, G., Rzepecki, J., Georgescu, R., Momennejad, I., Bignell, D., Sun, M., Shaw, A., Costello, G., Jacob, M., Devlin, S., & Hofmann, K. (2022). "How humans perceive human-like behavior in video game navigation." In CHI Conference on Human Factors in Computing Systems Extended Abstracts (CHI) 2022. pdf
bio
I am a true believer in the transformative role that artificial intelligence (AI) will continue to play in solving humanity's most challenging and impactful problems. I am currently a senior machine learning (ML) engineer at The Coalition studio with the Xbox Games Technology Group (XGTG). My role at XGTG is focused on building, incubating, and effectively sharing ML solutions for advancing the state of the art in game experiences and game development practices across Microsoft and the industry. I have worked on both deep technical research for applications in animation generation as well as building products from the ground up using large generative AI models for applications in quality assurance and game testing.
Prior to my work at The Coalition, I was the head of ML at Resolution Games. There I led the effort to use AI and ML techniques to create revolutionary XR player experiences and transform our game development processes. I am extremely happy to share that our work on human-like, player-facing bots trained with deep reinforcement learning and imitation learning techniques was shipped in the games Racket Club and Home Sports. The first of their kind bots in these games were the result of pursuing deep technical research to advance both human-like decision-making and behaviour for our players as well as the expressive control over these learned behaviours that game designers needed for authorial control over the experiences they were designing - all while running locally, in-engine on a VR headset. Depending on the game and the specific problem within it, these advances included employing goal-conditioned models as game designer tools, hierarchical systems that combined IL and RL in novel ways to improve bot skill and challenge ratings, learned animation blending to preserve highly-stylized animator intent in generated machine learning behaviour, and much, much more. You can also read more about some of our earlier research to get a taste for my approach to using techniques like deep learning, deep reinforcement learning, and similar technologies here.
My previous work as a post-doctoral researcher at Microsoft Research Cambridge used a human-centered AI research methodology — applying a combination of design, human-computer interaction, and AI/machine learning methods — to explore questions around the combination of AI and creativity in game development. This research focused on studying the challenges designers and developers might face when adopting AI techniques like deep reinforcement learning and imitation learning to create engaging game agents (non-player characters and bots) in real-world commercial games. I then developed technical prototypes using goal-conditioned RL and human in the loop RL for addressing some of these challenges. For a (relatively) quick overview of this research and some of my research as a Ph.D. scholar, you can watch this talk.
I received a Ph.D. in Computer Science (2019) from the Georgia Institute of Technology (Atlanta, USA) with the Expressive Machinery Lab. My dissertation investigated the effects of creative arc negotiation — a novel real-time decision-making paradigm for improvisation between people and computers — on player experience within VR games for improvisational theatre. I have previously studied the application of human-computer co-creativity in problems ranging from improvisational dance and pretend play to music recommendation. I received an M.S. in Computer Science (2013) from the Georgia Institute of Technology and a B.E. in Computer Science Engineering (2011) from the Manipal Institute of Technology (Manipal, India).
Recent work is covered in my CV below.
